minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[BUGFIX]: using language-only should not reserve space for the vision encoder (#18011)

Browse Source 
Co-authored-by: root <root@ubuntu-nvidia.localdomain>
This commit is contained in:
Koushik Dutta
2026-02-01 14:49:00 -08:00
committed by GitHub
co-authored by root
parent 1acae30806
commit 993ec178ef
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/server_args.py
+1 -1
View File
@@ -1057,7 +1057,7 @@ class ServerArgs:
# Multimodal models need more memory for the image processing, # Multimodal models need more memory for the image processing,
# so we adjust the mem_fraction_static accordingly. # so we adjust the mem_fraction_static accordingly.
model_config = self.get_model_config() model_config = self.get_model_config()
if model_config.is_multimodal: if model_config.is_multimodal and not self.language_only:
self.adjust_mem_fraction_for_vlm(model_config) self.adjust_mem_fraction_for_vlm(model_config)
# If symm mem is enabled and prealloc size is not set, set it to 4GB # If symm mem is enabled and prealloc size is not set, set it to 4GB